Quiz: AI Data Center Operations & Lifecycle (Part 14)

10 questions from Part 14 of the guide. Score instantly; every explanation links to the chapter that derives the answer.

Which reporting pair separates continuity from useful work in an AI fleet?
ticket closure ratecontracted service availability plus workload goodput, with separate denominatorsfacility availability (the nines)PUE alone
The DCIM/telemetry decision that makes failures diagnosable is…
buying more dashboardsalert de-duplication onlylonger retentioncorrelating facility (BMS/SCADA) and IT (DCGM/Redfish) data with synchronized time, quality flags and asset identities
The checkpoint tier mix exists to buy down…
model qualitymean time to recovery — RAM/peer tiers restore in seconds-to-minutes vs the persistent tier's longer pathstorage costnetwork bandwidth
Which inputs are required to set a spares service level for a failure class?
vendor recommendations onlyfleet population, demand/failure distribution, repair and replenishment distributions, target service level, common modes and available restoration pathsa fixed 5% rulewhatever fits the storeroom
How should POLCA’s reported training power headroom be used?
~3% in the evaluated training profile; measure the proposed workload before overbooking~50% — plentyunlimited with software caps~21% like inference
Firmware across the fleet (BMC, GPU, NIC, NVSwitch, PSU) should be treated as…
part of the attested golden stack, upgraded as a coordinated fleet event with canariesirrelevant to reliabilityeach vendor's auto-updater's jobset-and-forget
What triggers a refresh for a fleet slice, in the guide's framing?
calendar, efficiency (perf-per-watt vs the power budget), failure-rate or density-ramp signals — an explicit trigger you pick per slicewhen the vendor launches marketingthe calendar alonefailure of the whole fleet
The facility's end state should be decided as…
demolition, alwaysconversion to officesabandonment in placerepower vs demolish vs remediate — with repower keeping the shell and the hard-won megawatts for the next build
The 'autonomy ladder' for facility control loops runs…
local → cloudmanual → automaticobserve → recommend → act-within-envelope → fully autonomous, chosen per control loopoff → on
Disciplined operation on operating plant means…
outsourcing operationstrusting experienced techs' judgmentmatching SOP/MOP/switching authority to the action, with safe-state checks and explicit HOLD criteriafreezing all changes forever

Check your understanding: Day-2 Operations, Upgrades & Lifecycle